Are you talking to me? The rise of chatbots in education

By Sjoerd Stoffels

The latest buzz word in the world of education is ChatGPT. This type of artificial intelligence (AI) produces credible texts in a matter of seconds that are often usable for book reports, essays, and other assignments. The technology was made publicly available in November 2022 and millions are now experimenting with it, including students and academics. How should we interpret this technology and how can education deal with it? This blog post will discuss the essentials. FASoS staff can find more in-depth information in UMployees’s FASoS Education group.

Copyright: Laurent T. – Stock Illustration ID 561931702

ChatGPT: What are we talking about?

ChatGPT is an OpenAI Large Language Model that operates on the most recent version of GPT-3.5, where GPT stands for Generative Pre-trained Transformer. This refers to Large Language Models (LLMs) that can process entire sentences simultaneously, as opposed to word-by-word. This allows LLMs to process more data much faster, where Natural Language Processing (NLP) also plays a role in enabling the technology to understand, interpret and reproduce human language. LLMs can generate very plausible and coherent text due to being trained on vast sets of data, harvested from search engine indexes, Wikipedia, e-books, e-journals, etc. LLMs and their algorithms are already operating in search engines, social media, online reviews, virtual assistants, auto-completion, translation and writing assistants. GPT-based LLMs are common as well in the field of copywriting, translating, marketing, web design, coding and more. After ChatGPT’s worldwide public launch, it now has taken the educational field by storm too.

LLMs and their impact on education

The use of LLMs can help students hone their critical thinking and communication skills. To prevent plagiarism/fraud, assignments should be designed to be too complicated for these models to complete. The instructional design should challenge students to evaluate chatbot responses and improve them, rather than seeing LLMs as rivals to human intellect. In-class assignments, group work, and oral exams are preferred over take-home, open-book assignments. Chatbot models can be used in tutor group meetings for brainstorming and drafting initial essays. The Brookings Institution notes that LLMs only become a threat if the education system prioritises rubric points over knowledge and approaches students as knowledge digesters instead of transformers. Therefore, LLMs can serve as educational aids rather than enemies in teaching and learning.

LLM flaws and concerns

ChatGPT, like other LLMs, faces the problem of bias because it replicates the structure and themes present in its training data. The model’s generated text sounds plausible, but the user must verify the factual accuracy of the information conveyed. LLMs can also “hallucinate” by making up information, as it is trained to autocomplete generated output. While the model can provide good output on a generic level, the output will be mediocre for specific and complicated topics. In academic writing, attribution is important, but a LLM can only cite sources with a specific command. The model struggles to distinguish classic articles that must be cited from other articles that review the same content. It tends to keep referencing the same sources repeatedly, and it is more of a synthesizer than a critical thinker. There is also the risk that training and hosting of LLMs and its data will be limited to Big Tech corporations, due to their sheer size.

Tools for detecting AI-generated text

Considering that rewritten AI-generated text will be more difficult to detect, a suspected text can be submitted to stand-alone tools such as:

AI-generated text detection technology will at some point be integrated in the UM’s plagiarism service. The current service, Ouriginal, is not suited for such integration and will be replaced by the more potent Turnitin. Furthermore, automated watermarking of LLM generated output can be detected faster by stand-alone tools and plagiarism applications. This feature is not bulletproof yet and currently under development by OpenAI.

What’s next?

One doesn’t need a crystal ball to predict that ChatGPT is more than a hype and is here to stay. There will be more LLMs that can do a similar job; think of Alphabet’s Bard and Meta’s Galactica. Furthermore, LLMs are going to be integrated in search engines (GPT in Bing, Bard in Google), giving another push to everyday use of LLMs. Nevertheless, this type of AI is not yet that revolutionary as it is portrayed in the Sci-Fi and futuristic genre. ChatGPT and its associates still don’t pass a Turing or a Voight-Kampff test.

HAL (Heuristically programmed ALgorithmic) from 2001: A Space Odyssey’ plus copyrights (Copyright: User:Cryteria, CC BY 3.0, via Wikimedia Commons)

There are other emerging technologies with the potential to transform or perhaps even disrupt teaching and learning. Think of independent online learning platforms that can democratise education but also threaten the ‘business model’ of traditional higher education. Another one is the Open Educational Resources (OERs), which permit no-cost access, reuse, and adaptation of learning, teaching, and research materials. Again, this can democratise higher education and threaten its business model at the same time. Then we have Extended Reality (XR) technologies, which will immerse students and teachers in virtual environments, gaming, augmented or mixed reality. Finally, Neurotechnology and Brain-Machine Interfaces can link the human brain directly to databases, networks, and digital devices. All these new developments will influence the way students learn and teachers teach. This means that digital competences are crucial for teaching and learning. However, not just for the benefit of educational processes. Considering that these skills more and more prevail in a modern and increasing digital society.

About the author

Sjoerd Stoffels is project leader and consultant educational technology at FASoS. He has long-term experience in this domain, also being an eyewitness of its genesis. Sjoerd is active in several faculty, university and (inter)national educational technology proceedings. He was awarded with the UM Education Prize once, plus nominated two more times.